Fine tuning a model for customer service for our specific app

I don’t know a lot, but I can tell you that following the standard fine-tuning instructions here, OpenAI API, the model rejected the 3 keys (prompt, completion, label) above. So I submitted the dataset with just the prompt and completion. Results were varied. Everyone keeps telling me that to get the best results (answers that match your dataset more closely) you have to use “embeddings”. Trying to learn what that is and how to do it now. See: Google Spreadsheet + GPT3